Using machine learning and the first digit law to detect forgeries in digital images

Abstract

Digital image tampering is becoming popular and it might cause serious problems in different areas. Therefore, detection forgeries in digital images are an urgent need. There are various forgery types, which can be exposed by different forensic techniques. In this paper, we propose a new detection scheme using the first-digit law (also known as Benford's law) in order to identify several types of image forgeries. We extract specific features, which are fed to a machine learning based classifier in order to distinguish between original images and manipulated images. Through experiments, we found that the proposed scheme works well for detecting double JPEG compression and Gaussian noise addition. Copy-move is among the most popular types of image forgeries, where a part of an image is copied and pasted to another position of the same image. However, we show this manipulation does not affect the law. Experiments on a large-scale image dataset show that the proposed scheme is reliable and it can achieve detection rate up to 90% or higher.